CERN Accelerating science

Previous:		Clearance of Building 186 - Tape Archive	(See printing version)
Next:		Advanced Warning for VMS System Managers ...

Tapes....

If you want to store a lot of data it is still probable that you will need to store the majority of it on tape of some sort. Disk storage at ~100 CHF/GByte for TB amounts is still considerably more than the lowest cost tape we can use today at CERN (STK 50 GB Redwood) which approaches 1 CHF/GB, assuming a large purchase of tape and the typical CERN compression factor seen on such tapes.

The major LHC experiments are currently expected to collect about 4 Petabytes (PB) of raw data every year. With today's tape systems, and allowing for import /export of data and collections of analysed events, we might expect to see a demand for perhaps 100,000 new tapes every year. If the size of an experimental 'data file' is assumed to be 1 Gigabyte, this also implies writing several million files on tape every year. How can all this data be accessed in a reliable and timely fashion?

A part of the answer is that tape handling will be almost totally automated. Once in a robotic library, a tape is reasonably immune from typical human errors, which plague manual systems, such as tapes getting lost, being dropped or being improperly reinitialised. Once a tape is mounted on a unit, it is still important for the users to be confident that the correct data file is about to be read (or, even more importantly, about to be written).

It is also helpful, from the point of view of efficiency, when a tape can already contain more than 50 Gigabytes of data and thousands of files, to be able to position it as fast as possible to the correct data file. This can be done of course by reading all the way along the tape until the correct data is found, but, even at 10 MBytes/second, this might take an hour for a well-filled tape.....

Tape labels are another part of the answer....

The use of tape labels has been strongly recommended for many years at CERN by IT. Suitable tape labels permit an automated library to accurately and repeatedly select the correct tape for a task. Most major 'historic' operating systems such as CDC's Scope and NOS/BE, IBM's MVS and VM, or Cray's Unicos fully supported tape label use. They check that the correct tape has been mounted for a particular task, and obtain much of the information needed to process the tape itself directly from the labels. Use of such labels allows rapid positioning to a particular file on a tape, and some degree of protection against user errors can be provided.

...and the X3.27 ANSI standard helped

This relatively old ANSI standard defines a structure for labelled magnetic tapes and specifies the contents of the labels involved. It provides a widely accepted means of writing a labelled tape to exchange with another organisation (and possibly another operating system), where the labels again provide much of the information needed to allow the tape to be read again successfully. The labels on a tape also help ease the problems of reading them back on upgraded hardware, or under the control of a new operating system. After all, tapes have a life of 10-20 years, if well looked after, which is rather longer than most operating systems or hardware platforms that CERN has used so far.

However, recent developments in tape technology have made the X3.27 standard increasingly obsolete. This cannot be a great surprise: the original CERN Computer Science library copy of the IBM manual GC26-3795-3, 'OS/VS Tape Labels', which refers to 'American National Standard Magnetic Tape Labels for Information Interchange, X3.27-1969', was placed on the shelf on 12th March 1979. It is still a reasonably good guide as far as tape label use at CERN is concerned, but the comments made in it concerning tape drives have a slightly antique flavour.

The American National Standards Institute itself has not stood still: ANSI maintains a WWW site, of course, which you can find at:

This offers a search tool which will directly locate document number 'X3.27' for you:

You will unfortunately not find much information about X3.27 directly here, or any description of its evolution over time. You do, however, have an on-line option to purchase the standards document itself ('ANSI X3.27 - 1987 (R1993)'). I would have hoped at least for a hint about the most recent changes, to see if this was worth doing! You can also offer e-mail feedback to the organisation about X3.27.

Tape labels as used at CERN

For those unfamiliar with tape labels, I invite you to look at the detailed description made on this subject (general features, types of labels - VOL1, HDR1 and HDR2, EOF1 and EOF2 -, labelling software, scanning or dumping labels, etc.) at URL:

N.B. This is a "long version" for this article, that you could read as a cure for insomnia... It also describes how the X3.27 standard recommendations are made use of in tape label processing as practised at CERN. Some of the limitations we now see with existing tape labelling are noted, as well as some options for improvement.

In the next CNL, a summary of the stage commands and their options will appear, with a selection of examples relevant to normal tape use at CERN. This will supplement the information which is already available to you via the 'man' command (i.e. 'man stagein').

For matters related to this article please contact the author.